Google Buys Android App Code From Developers to Train Gemini

*Google is paying select Play Store developers for access to their private codebases under a confidential program aimed at improving its lagging AI coding tools.*

The move

Google has begun purchasing source code from Android developers who publish on the Play Store. The effort is framed internally as a way to gather high-quality, real-world examples that current public datasets lack.

Two reports published on the same day describe the same program. Neowin frames it as a proposal Google is making to developers because Gemini trails Claude Code and Copilot. 404 Media reports that Google is already executing purchases through a “confidential” channel and has approached multiple developers.

What developers are being offered

The company is not seeking open-source repositories that are already public. It wants private or semi-private codebases that reflect production Android apps. In return, developers receive payment whose size has not been disclosed.

No formal public announcement has been made. Outreach appears selective and is handled under nondisclosure terms, according to the reporting.

Why the purchases are happening now

Public code repositories have been scraped for years. The remaining gap for model training lies in the patterns found inside shipping commercial apps—error handling, library usage, and integration choices that rarely appear in clean open-source examples.

Google’s coding model Gemini has not closed the gap with the leading alternatives. The purchases represent a direct attempt to obtain the data distribution the model currently lacks.

Limits of the approach

Buying code does not automatically produce better models. The acquired repositories must still be filtered, deduplicated, and balanced against existing training data. Developers who sell their code also lose future control over how that code shapes downstream AI tools.

The program remains small and invitation-only. No evidence has surfaced that it extends beyond a narrow set of Play Store authors.

What it signals

Large labs have exhausted easy sources of training data. When a company starts writing checks for private code, it is acknowledging that scale alone is no longer sufficient and that quality, even at modest volumes, now carries a price.

---

Sources:

• Neowin
• 404 Media

{
  "excerpt": "Google is paying select Play Store developers for private Android codebases to improve Gemini's training data under a confidential program.",
  "suggestedSection": "ai",
  "suggestedTags": ["google", "gemini", "ai-training", "android"],
  "imagePrompt": "Abstract composition of scattered code printouts and circuit traces layered on a dark wooden table, soft overhead light casting long shadows across the paper edges. muted color palette, cinematic lighting, 16:9"
}